examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <unistd.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76 #include <rte_timer.h>
  77 #include <rte_power.h>
  78
  79 #include "main.h"
  80
  81 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  82
  83 #define MAX_PKT_BURST 32
  84
  85 #define MIN_ZERO_POLL_COUNT 5
  86
  87 /* around 100ms at 2 Ghz */
  88 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  89 /* 100 ms interval */
  90 #define TIMER_NUMBER_PER_SECOND           10
  91 /* 100000 us */
  92 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  93 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  94
  95 #define APP_LOOKUP_EXACT_MATCH          0
  96 #define APP_LOOKUP_LPM                  1
  97 #define DO_RFC_1812_CHECKS
  98
  99 #ifndef APP_LOOKUP_METHOD
 100 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
 101 #endif
 102
 103 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 104 #include <rte_hash.h>
 105 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 106 #include <rte_lpm.h>
 107 #else
 108 #error "APP_LOOKUP_METHOD set to incorrect value"
 109 #endif
 110
 111 #ifndef IPv6_BYTES
 112 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 113                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 114 #define IPv6_BYTES(addr) \
 115         addr[0],  addr[1], addr[2],  addr[3], \
 116         addr[4],  addr[5], addr[6],  addr[7], \
 117         addr[8],  addr[9], addr[10], addr[11],\
 118         addr[12], addr[13],addr[14], addr[15]
 119 #endif
 120
 121 #define MAX_JUMBO_PKT_LEN  9600
 122
 123 #define IPV6_ADDR_LEN 16
 124
 125 #define MEMPOOL_CACHE_SIZE 256
 126
 127 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 128
 129 /*
 130  * This expression is used to calculate the number of mbufs needed depending on
 131  * user input, taking into account memory for rx and tx hardware rings, cache
 132  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 133  * NB_MBUF never goes below a minimum value of 8192.
 134  */
 135
 136 #define NB_MBUF RTE_MAX ( \
 137         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 138         nb_ports*nb_lcores*MAX_PKT_BURST + \
 139         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 140         nb_lcores*MEMPOOL_CACHE_SIZE), \
 141         (unsigned)8192)
 142
 143 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 144
 145 #define NB_SOCKETS 8
 146
 147 /* Configure how many packets ahead to prefetch, when reading packets */
 148 #define PREFETCH_OFFSET 3
 149
 150 /*
 151  * Configurable number of RX/TX ring descriptors
 152  */
 153 #define RTE_TEST_RX_DESC_DEFAULT 128
 154 #define RTE_TEST_TX_DESC_DEFAULT 512
 155 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 156 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 157
 158 /* ethernet addresses of ports */
 159 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 160
 161 /* mask of enabled ports */
 162 static uint32_t enabled_port_mask = 0;
 163 /* Ports set in promiscuous mode off by default. */
 164 static int promiscuous_on = 0;
 165 /* NUMA is enabled by default. */
 166 static int numa_on = 1;
 167
 168 enum freq_scale_hint_t
 169 {
 170         FREQ_LOWER    =      -1,
 171         FREQ_CURRENT  =       0,
 172         FREQ_HIGHER   =       1,
 173         FREQ_HIGHEST  =       2
 174 };
 175
 176 struct mbuf_table {
 177         uint16_t len;
 178         struct rte_mbuf *m_table[MAX_PKT_BURST];
 179 };
 180
 181 struct lcore_rx_queue {
 182         uint8_t port_id;
 183         uint8_t queue_id;
 184         enum freq_scale_hint_t freq_up_hint;
 185         uint32_t zero_rx_packet_count;
 186         uint32_t idle_hint;
 187 } __rte_cache_aligned;
 188
 189 #define MAX_RX_QUEUE_PER_LCORE 16
 190 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 191 #define MAX_RX_QUEUE_PER_PORT 128
 192
 193 #define MAX_LCORE_PARAMS 1024
 194 struct lcore_params {
 195         uint8_t port_id;
 196         uint8_t queue_id;
 197         uint8_t lcore_id;
 198 } __rte_cache_aligned;
 199
 200 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 201 static struct lcore_params lcore_params_array_default[] = {
 202         {0, 0, 2},
 203         {0, 1, 2},
 204         {0, 2, 2},
 205         {1, 0, 2},
 206         {1, 1, 2},
 207         {1, 2, 2},
 208         {2, 0, 2},
 209         {3, 0, 3},
 210         {3, 1, 3},
 211 };
 212
 213 static struct lcore_params * lcore_params = lcore_params_array_default;
 214 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 215                                 sizeof(lcore_params_array_default[0]);
 216
 217 static struct rte_eth_conf port_conf = {
 218         .rxmode = {
 219                 .mq_mode        = ETH_MQ_RX_RSS,
 220                 .max_rx_pkt_len = ETHER_MAX_LEN,
 221                 .split_hdr_size = 0,
 222                 .header_split   = 0, /**< Header Split disabled */
 223                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 224                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 225                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 226                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 227         },
 228         .rx_adv_conf = {
 229                 .rss_conf = {
 230                         .rss_key = NULL,
 231                         .rss_hf = ETH_RSS_IP,
 232                 },
 233         },
 234         .txmode = {
 235                 .mq_mode = ETH_DCB_NONE,
 236         },
 237 };
 238
 239 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 240
 241
 242 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 243
 244 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 245 #include <rte_hash_crc.h>
 246 #define DEFAULT_HASH_FUNC       rte_hash_crc
 247 #else
 248 #include <rte_jhash.h>
 249 #define DEFAULT_HASH_FUNC       rte_jhash
 250 #endif
 251
 252 struct ipv4_5tuple {
 253         uint32_t ip_dst;
 254         uint32_t ip_src;
 255         uint16_t port_dst;
 256         uint16_t port_src;
 257         uint8_t  proto;
 258 } __attribute__((__packed__));
 259
 260 struct ipv6_5tuple {
 261         uint8_t  ip_dst[IPV6_ADDR_LEN];
 262         uint8_t  ip_src[IPV6_ADDR_LEN];
 263         uint16_t port_dst;
 264         uint16_t port_src;
 265         uint8_t  proto;
 266 } __attribute__((__packed__));
 267
 268 struct ipv4_l3fwd_route {
 269         struct ipv4_5tuple key;
 270         uint8_t if_out;
 271 };
 272
 273 struct ipv6_l3fwd_route {
 274         struct ipv6_5tuple key;
 275         uint8_t if_out;
 276 };
 277
 278 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 279         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 280         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 281         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 282         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 283 };
 284
 285 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 286         {
 287                 {
 288                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 289                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 290                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 291                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 292                          1, 10, IPPROTO_UDP
 293                 }, 4
 294         },
 295 };
 296
 297 typedef struct rte_hash lookup_struct_t;
 298 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 299 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 300
 301 #define L3FWD_HASH_ENTRIES      1024
 302
 303 #define IPV4_L3FWD_NUM_ROUTES \
 304         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 305
 306 #define IPV6_L3FWD_NUM_ROUTES \
 307         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 308
 309 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 310 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 311 #endif
 312
 313 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 314 struct ipv4_l3fwd_route {
 315         uint32_t ip;
 316         uint8_t  depth;
 317         uint8_t  if_out;
 318 };
 319
 320 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 321         {IPv4(1,1,1,0), 24, 0},
 322         {IPv4(2,1,1,0), 24, 1},
 323         {IPv4(3,1,1,0), 24, 2},
 324         {IPv4(4,1,1,0), 24, 3},
 325         {IPv4(5,1,1,0), 24, 4},
 326         {IPv4(6,1,1,0), 24, 5},
 327         {IPv4(7,1,1,0), 24, 6},
 328         {IPv4(8,1,1,0), 24, 7},
 329 };
 330
 331 #define IPV4_L3FWD_NUM_ROUTES \
 332         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 333
 334 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 335
 336 typedef struct rte_lpm lookup_struct_t;
 337 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 338 #endif
 339
 340 struct lcore_conf {
 341         uint16_t n_rx_queue;
 342         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 343         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 344         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 345         lookup_struct_t * ipv4_lookup_struct;
 346         lookup_struct_t * ipv6_lookup_struct;
 347 } __rte_cache_aligned;
 348
 349 struct lcore_stats {
 350         /* total sleep time in ms since last frequency scaling down */
 351         uint32_t sleep_time;
 352         /* number of long sleep recently */
 353         uint32_t nb_long_sleep;
 354         /* freq. scaling up trend */
 355         uint32_t trend;
 356         /* total packet processed recently */
 357         uint64_t nb_rx_processed;
 358         /* total iterations looped recently */
 359         uint64_t nb_iteration_looped;
 360         uint32_t padding[9];
 361 } __rte_cache_aligned;
 362
 363 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 364 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 365 static struct rte_timer power_timers[RTE_MAX_LCORE];
 366
 367 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 368 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 369                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 370
 371 /* exit signal handler */
 372 static void
 373 signal_exit_now(int sigtype)
 374 {
 375         unsigned lcore_id;
 376         int ret;
 377
 378         if (sigtype == SIGINT) {
 379                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 380                         if (rte_lcore_is_enabled(lcore_id) == 0)
 381                                 continue;
 382
 383                         /* init power management library */
 384                         ret = rte_power_exit(lcore_id);
 385                         if (ret)
 386                                 rte_exit(EXIT_FAILURE, "Power management "
 387                                         "library de-initialization failed on "
 388                                                         "core%u\n", lcore_id);
 389                 }
 390         }
 391
 392         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 393 }
 394
 395 /*  Freqency scale down timer callback */
 396 static void
 397 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 398                           __attribute__((unused)) void *arg)
 399 {
 400         uint64_t hz;
 401         float sleep_time_ratio;
 402         unsigned lcore_id = rte_lcore_id();
 403
 404         /* accumulate total execution time in us when callback is invoked */
 405         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 406                                         (float)SCALING_PERIOD;
 407
 408         /**
 409          * check whether need to scale down frequency a step if it sleep a lot.
 410          */
 411         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 412                 rte_power_freq_down(lcore_id);
 413         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 414                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 415                 /**
 416                  * scale down a step if average packet per iteration less
 417                  * than expectation.
 418                  */
 419                 rte_power_freq_down(lcore_id);
 420
 421         /**
 422          * initialize another timer according to current frequency to ensure
 423          * timer interval is relatively fixed.
 424          */
 425         hz = rte_get_timer_hz();
 426         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 427                                 SINGLE, lcore_id, power_timer_cb, NULL);
 428
 429         stats[lcore_id].nb_rx_processed = 0;
 430         stats[lcore_id].nb_iteration_looped = 0;
 431
 432         stats[lcore_id].sleep_time = 0;
 433 }
 434
 435 /* Send burst of packets on an output interface */
 436 static inline int
 437 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 438 {
 439         struct rte_mbuf **m_table;
 440         int ret;
 441         uint16_t queueid;
 442
 443         queueid = qconf->tx_queue_id[port];
 444         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 445
 446         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 447         if (unlikely(ret < n)) {
 448                 do {
 449                         rte_pktmbuf_free(m_table[ret]);
 450                 } while (++ret < n);
 451         }
 452
 453         return 0;
 454 }
 455
 456 /* Enqueue a single packet, and send burst if queue is filled */
 457 static inline int
 458 send_single_packet(struct rte_mbuf *m, uint8_t port)
 459 {
 460         uint32_t lcore_id;
 461         uint16_t len;
 462         struct lcore_conf *qconf;
 463
 464         lcore_id = rte_lcore_id();
 465
 466         qconf = &lcore_conf[lcore_id];
 467         len = qconf->tx_mbufs[port].len;
 468         qconf->tx_mbufs[port].m_table[len] = m;
 469         len++;
 470
 471         /* enough pkts to be sent */
 472         if (unlikely(len == MAX_PKT_BURST)) {
 473                 send_burst(qconf, MAX_PKT_BURST, port);
 474                 len = 0;
 475         }
 476
 477         qconf->tx_mbufs[port].len = len;
 478         return 0;
 479 }
 480
 481 #ifdef DO_RFC_1812_CHECKS
 482 static inline int
 483 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 484 {
 485         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 486         /*
 487          * 1. The packet length reported by the Link Layer must be large
 488          * enough to hold the minimum length legal IP datagram (20 bytes).
 489          */
 490         if (link_len < sizeof(struct ipv4_hdr))
 491                 return -1;
 492
 493         /* 2. The IP checksum must be correct. */
 494         /* this is checked in H/W */
 495
 496         /*
 497          * 3. The IP version number must be 4. If the version number is not 4
 498          * then the packet may be another version of IP, such as IPng or
 499          * ST-II.
 500          */
 501         if (((pkt->version_ihl) >> 4) != 4)
 502                 return -3;
 503         /*
 504          * 4. The IP header length field must be large enough to hold the
 505          * minimum length legal IP datagram (20 bytes = 5 words).
 506          */
 507         if ((pkt->version_ihl & 0xf) < 5)
 508                 return -4;
 509
 510         /*
 511          * 5. The IP total length field must be large enough to hold the IP
 512          * datagram header, whose length is specified in the IP header length
 513          * field.
 514          */
 515         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 516                 return -5;
 517
 518         return 0;
 519 }
 520 #endif
 521
 522 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 523 static void
 524 print_ipv4_key(struct ipv4_5tuple key)
 525 {
 526         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 527                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 528                                 key.port_dst, key.port_src, key.proto);
 529 }
 530 static void
 531 print_ipv6_key(struct ipv6_5tuple key)
 532 {
 533         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 534                 "port dst = %d, port src = %d, proto = %d\n",
 535                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 536                 key.port_dst, key.port_src, key.proto);
 537 }
 538
 539 static inline uint8_t
 540 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 541                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 542 {
 543         struct ipv4_5tuple key;
 544         struct tcp_hdr *tcp;
 545         struct udp_hdr *udp;
 546         int ret = 0;
 547
 548         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 549         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 550         key.proto = ipv4_hdr->next_proto_id;
 551
 552         switch (ipv4_hdr->next_proto_id) {
 553         case IPPROTO_TCP:
 554                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 555                                         sizeof(struct ipv4_hdr));
 556                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 557                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 558                 break;
 559
 560         case IPPROTO_UDP:
 561                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 562                                         sizeof(struct ipv4_hdr));
 563                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 564                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 565                 break;
 566
 567         default:
 568                 key.port_dst = 0;
 569                 key.port_src = 0;
 570                 break;
 571         }
 572
 573         /* Find destination port */
 574         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 575         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 576 }
 577
 578 static inline uint8_t
 579 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 580                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 581 {
 582         struct ipv6_5tuple key;
 583         struct tcp_hdr *tcp;
 584         struct udp_hdr *udp;
 585         int ret = 0;
 586
 587         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 588         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 589
 590         key.proto = ipv6_hdr->proto;
 591
 592         switch (ipv6_hdr->proto) {
 593         case IPPROTO_TCP:
 594                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 595                                         sizeof(struct ipv6_hdr));
 596                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 597                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 598                 break;
 599
 600         case IPPROTO_UDP:
 601                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 602                                         sizeof(struct ipv6_hdr));
 603                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 604                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 605                 break;
 606
 607         default:
 608                 key.port_dst = 0;
 609                 key.port_src = 0;
 610                 break;
 611         }
 612
 613         /* Find destination port */
 614         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 615         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 616 }
 617 #endif
 618
 619 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 620 static inline uint8_t
 621 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 622                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 623 {
 624         uint8_t next_hop;
 625
 626         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 627                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 628                         next_hop : portid);
 629 }
 630 #endif
 631
 632 static inline void
 633 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 634                                 struct lcore_conf *qconf)
 635 {
 636         struct ether_hdr *eth_hdr;
 637         struct ipv4_hdr *ipv4_hdr;
 638         void *d_addr_bytes;
 639         uint8_t dst_port;
 640
 641         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 642
 643         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 644                 /* Handle IPv4 headers.*/
 645                 ipv4_hdr =
 646                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 647                                                 + sizeof(struct ether_hdr));
 648
 649 #ifdef DO_RFC_1812_CHECKS
 650                 /* Check to make sure the packet is valid (RFC1812) */
 651                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 652                         rte_pktmbuf_free(m);
 653                         return;
 654                 }
 655 #endif
 656
 657                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 658                                         qconf->ipv4_lookup_struct);
 659                 if (dst_port >= RTE_MAX_ETHPORTS ||
 660                                 (enabled_port_mask & 1 << dst_port) == 0)
 661                         dst_port = portid;
 662
 663                 /* 02:00:00:00:00:xx */
 664                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 665                 *((uint64_t *)d_addr_bytes) =
 666                         0x000000000002 + ((uint64_t)dst_port << 40);
 667
 668 #ifdef DO_RFC_1812_CHECKS
 669                 /* Update time to live and header checksum */
 670                 --(ipv4_hdr->time_to_live);
 671                 ++(ipv4_hdr->hdr_checksum);
 672 #endif
 673
 674                 /* src addr */
 675                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 676
 677                 send_single_packet(m, dst_port);
 678         }
 679         else {
 680                 /* Handle IPv6 headers.*/
 681 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 682                 struct ipv6_hdr *ipv6_hdr;
 683
 684                 ipv6_hdr =
 685                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 686                                                 + sizeof(struct ether_hdr));
 687
 688                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 689                                         qconf->ipv6_lookup_struct);
 690
 691                 if (dst_port >= RTE_MAX_ETHPORTS ||
 692                                 (enabled_port_mask & 1 << dst_port) == 0)
 693                         dst_port = portid;
 694
 695                 /* 02:00:00:00:00:xx */
 696                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 697                 *((uint64_t *)d_addr_bytes) =
 698                         0x000000000002 + ((uint64_t)dst_port << 40);
 699
 700                 /* src addr */
 701                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 702
 703                 send_single_packet(m, dst_port);
 704 #else
 705                 /* We don't currently handle IPv6 packets in LPM mode. */
 706                 rte_pktmbuf_free(m);
 707 #endif
 708         }
 709
 710 }
 711
 712 #define SLEEP_GEAR1_THRESHOLD            100
 713 #define SLEEP_GEAR2_THRESHOLD            1000
 714
 715 static inline uint32_t
 716 power_idle_heuristic(uint32_t zero_rx_packet_count)
 717 {
 718         /* If zero count is less than 100, use it as the sleep time in us */
 719         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 720                 return zero_rx_packet_count;
 721         /* If zero count is less than 1000, sleep time should be 100 us */
 722         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 723                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 724                 return SLEEP_GEAR1_THRESHOLD;
 725         /* If zero count is greater than 1000, sleep time should be 1000 us */
 726         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 727                 return SLEEP_GEAR2_THRESHOLD;
 728
 729         return 0;
 730 }
 731
 732 static inline enum freq_scale_hint_t
 733 power_freq_scaleup_heuristic(unsigned lcore_id,
 734                              uint8_t port_id,
 735                              uint16_t queue_id)
 736 {
 737 /**
 738  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 739  * per iteration
 740  */
 741 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 742 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 743 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 744 #define FREQ_UP_TREND1_ACC   1
 745 #define FREQ_UP_TREND2_ACC   100
 746 #define FREQ_UP_THRESHOLD    10000
 747
 748         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 749                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 750                 stats[lcore_id].trend = 0;
 751                 return FREQ_HIGHEST;
 752         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 753                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 754                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 755         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 756                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 757                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 758
 759         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 760                 stats[lcore_id].trend = 0;
 761                 return FREQ_HIGHER;
 762         }
 763
 764         return FREQ_CURRENT;
 765 }
 766
 767 /* main processing loop */
 768 static int
 769 main_loop(__attribute__((unused)) void *dummy)
 770 {
 771         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 772         unsigned lcore_id;
 773         uint64_t prev_tsc, diff_tsc, cur_tsc;
 774         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 775         int i, j, nb_rx;
 776         uint8_t portid, queueid;
 777         struct lcore_conf *qconf;
 778         struct lcore_rx_queue *rx_queue;
 779         enum freq_scale_hint_t lcore_scaleup_hint;
 780
 781         uint32_t lcore_rx_idle_count = 0;
 782         uint32_t lcore_idle_hint = 0;
 783
 784         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 785
 786         prev_tsc = 0;
 787
 788         lcore_id = rte_lcore_id();
 789         qconf = &lcore_conf[lcore_id];
 790
 791         if (qconf->n_rx_queue == 0) {
 792                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 793                 return 0;
 794         }
 795
 796         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 797
 798         for (i = 0; i < qconf->n_rx_queue; i++) {
 799
 800                 portid = qconf->rx_queue_list[i].port_id;
 801                 queueid = qconf->rx_queue_list[i].queue_id;
 802                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 803                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 804         }
 805
 806         while (1) {
 807                 stats[lcore_id].nb_iteration_looped++;
 808
 809                 cur_tsc = rte_rdtsc();
 810                 cur_tsc_power = cur_tsc;
 811
 812                 /*
 813                  * TX burst queue drain
 814                  */
 815                 diff_tsc = cur_tsc - prev_tsc;
 816                 if (unlikely(diff_tsc > drain_tsc)) {
 817
 818                         /*
 819                          * This could be optimized (use queueid instead of
 820                          * portid), but it is not called so often
 821                          */
 822                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 823                                 if (qconf->tx_mbufs[portid].len == 0)
 824                                         continue;
 825                                 send_burst(&lcore_conf[lcore_id],
 826                                         qconf->tx_mbufs[portid].len,
 827                                         portid);
 828                                 qconf->tx_mbufs[portid].len = 0;
 829                         }
 830
 831                         prev_tsc = cur_tsc;
 832                 }
 833
 834                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 835                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 836                         rte_timer_manage();
 837                         prev_tsc_power = cur_tsc_power;
 838                 }
 839
 840                 /*
 841                  * Read packet from RX queues
 842                  */
 843                 lcore_scaleup_hint = FREQ_CURRENT;
 844                 lcore_rx_idle_count = 0;
 845                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 846                         rx_queue = &(qconf->rx_queue_list[i]);
 847                         rx_queue->idle_hint = 0;
 848                         portid = rx_queue->port_id;
 849                         queueid = rx_queue->queue_id;
 850
 851                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 852                                                                 MAX_PKT_BURST);
 853                         stats[lcore_id].nb_rx_processed += nb_rx;
 854                         if (unlikely(nb_rx == 0)) {
 855                                 /**
 856                                  * no packet received from rx queue, try to
 857                                  * sleep for a while forcing CPU enter deeper
 858                                  * C states.
 859                                  */
 860                                 rx_queue->zero_rx_packet_count++;
 861
 862                                 if (rx_queue->zero_rx_packet_count <=
 863                                                         MIN_ZERO_POLL_COUNT)
 864                                         continue;
 865
 866                                 rx_queue->idle_hint = power_idle_heuristic(\
 867                                         rx_queue->zero_rx_packet_count);
 868                                 lcore_rx_idle_count++;
 869                         } else {
 870                                 rx_queue->zero_rx_packet_count = 0;
 871
 872                                 /**
 873                                  * do not scale up frequency immediately as
 874                                  * user to kernel space communication is costly
 875                                  * which might impact packet I/O for received
 876                                  * packets.
 877                                  */
 878                                 rx_queue->freq_up_hint =
 879                                         power_freq_scaleup_heuristic(lcore_id,
 880                                                         portid, queueid);
 881                         }
 882
 883                         /* Prefetch first packets */
 884                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 885                                 rte_prefetch0(rte_pktmbuf_mtod(
 886                                                 pkts_burst[j], void *));
 887                         }
 888
 889                         /* Prefetch and forward already prefetched packets */
 890                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 891                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 892                                                 j + PREFETCH_OFFSET], void *));
 893                                 l3fwd_simple_forward(pkts_burst[j], portid,
 894                                                                 qconf);
 895                         }
 896
 897                         /* Forward remaining prefetched packets */
 898                         for (; j < nb_rx; j++) {
 899                                 l3fwd_simple_forward(pkts_burst[j], portid,
 900                                                                 qconf);
 901                         }
 902                 }
 903
 904                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 905                         for (i = 1, lcore_scaleup_hint =
 906                                 qconf->rx_queue_list[0].freq_up_hint;
 907                                         i < qconf->n_rx_queue; ++i) {
 908                                 rx_queue = &(qconf->rx_queue_list[i]);
 909                                 if (rx_queue->freq_up_hint >
 910                                                 lcore_scaleup_hint)
 911                                         lcore_scaleup_hint =
 912                                                 rx_queue->freq_up_hint;
 913                         }
 914
 915                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 916                                 rte_power_freq_max(lcore_id);
 917                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 918                                 rte_power_freq_up(lcore_id);
 919                 } else {
 920                         /**
 921                          * All Rx queues empty in recent consecutive polls,
 922                          * sleep in a conservative manner, meaning sleep as
 923                          * less as possible.
 924                          */
 925                         for (i = 1, lcore_idle_hint =
 926                                 qconf->rx_queue_list[0].idle_hint;
 927                                         i < qconf->n_rx_queue; ++i) {
 928                                 rx_queue = &(qconf->rx_queue_list[i]);
 929                                 if (rx_queue->idle_hint < lcore_idle_hint)
 930                                         lcore_idle_hint = rx_queue->idle_hint;
 931                         }
 932
 933                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 934                                 /**
 935                                  * execute "pause" instruction to avoid context
 936                                  * switch for short sleep.
 937                                  */
 938                                 rte_delay_us(lcore_idle_hint);
 939                         else
 940                                 /* long sleep force runing thread to suspend */
 941                                 usleep(lcore_idle_hint);
 942
 943                         stats[lcore_id].sleep_time += lcore_idle_hint;
 944                 }
 945         }
 946 }
 947
 948 static int
 949 check_lcore_params(void)
 950 {
 951         uint8_t queue, lcore;
 952         uint16_t i;
 953         int socketid;
 954
 955         for (i = 0; i < nb_lcore_params; ++i) {
 956                 queue = lcore_params[i].queue_id;
 957                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 958                         printf("invalid queue number: %hhu\n", queue);
 959                         return -1;
 960                 }
 961                 lcore = lcore_params[i].lcore_id;
 962                 if (!rte_lcore_is_enabled(lcore)) {
 963                         printf("error: lcore %hhu is not enabled in lcore "
 964                                                         "mask\n", lcore);
 965                         return -1;
 966                 }
 967                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 968                                                         (numa_on == 0)) {
 969                         printf("warning: lcore %hhu is on socket %d with numa "
 970                                                 "off\n", lcore, socketid);
 971                 }
 972         }
 973         return 0;
 974 }
 975
 976 static int
 977 check_port_config(const unsigned nb_ports)
 978 {
 979         unsigned portid;
 980         uint16_t i;
 981
 982         for (i = 0; i < nb_lcore_params; ++i) {
 983                 portid = lcore_params[i].port_id;
 984                 if ((enabled_port_mask & (1 << portid)) == 0) {
 985                         printf("port %u is not enabled in port mask\n",
 986                                                                 portid);
 987                         return -1;
 988                 }
 989                 if (portid >= nb_ports) {
 990                         printf("port %u is not present on the board\n",
 991                                                                 portid);
 992                         return -1;
 993                 }
 994         }
 995         return 0;
 996 }
 997
 998 static uint8_t
 999 get_port_n_rx_queues(const uint8_t port)
1000 {
1001         int queue = -1;
1002         uint16_t i;
1003
1004         for (i = 0; i < nb_lcore_params; ++i) {
1005                 if (lcore_params[i].port_id == port &&
1006                                 lcore_params[i].queue_id > queue)
1007                         queue = lcore_params[i].queue_id;
1008         }
1009         return (uint8_t)(++queue);
1010 }
1011
1012 static int
1013 init_lcore_rx_queues(void)
1014 {
1015         uint16_t i, nb_rx_queue;
1016         uint8_t lcore;
1017
1018         for (i = 0; i < nb_lcore_params; ++i) {
1019                 lcore = lcore_params[i].lcore_id;
1020                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1021                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1022                         printf("error: too many queues (%u) for lcore: %u\n",
1023                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1024                         return -1;
1025                 } else {
1026                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1027                                 lcore_params[i].port_id;
1028                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1029                                 lcore_params[i].queue_id;
1030                         lcore_conf[lcore].n_rx_queue++;
1031                 }
1032         }
1033         return 0;
1034 }
1035
1036 /* display usage */
1037 static void
1038 print_usage(const char *prgname)
1039 {
1040         printf ("%s [EAL options] -- -p PORTMASK -P"
1041                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1042                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1043                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1044                 "  -P : enable promiscuous mode\n"
1045                 "  --config (port,queue,lcore): rx queues configuration\n"
1046                 "  --no-numa: optional, disable numa awareness\n"
1047                 "  --enable-jumbo: enable jumbo frame"
1048                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1049                 prgname);
1050 }
1051
1052 static int parse_max_pkt_len(const char *pktlen)
1053 {
1054         char *end = NULL;
1055         unsigned long len;
1056
1057         /* parse decimal string */
1058         len = strtoul(pktlen, &end, 10);
1059         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1060                 return -1;
1061
1062         if (len == 0)
1063                 return -1;
1064
1065         return len;
1066 }
1067
1068 static int
1069 parse_portmask(const char *portmask)
1070 {
1071         char *end = NULL;
1072         unsigned long pm;
1073
1074         /* parse hexadecimal string */
1075         pm = strtoul(portmask, &end, 16);
1076         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1077                 return -1;
1078
1079         if (pm == 0)
1080                 return -1;
1081
1082         return pm;
1083 }
1084
1085 static int
1086 parse_config(const char *q_arg)
1087 {
1088         char s[256];
1089         const char *p, *p0 = q_arg;
1090         char *end;
1091         enum fieldnames {
1092                 FLD_PORT = 0,
1093                 FLD_QUEUE,
1094                 FLD_LCORE,
1095                 _NUM_FLD
1096         };
1097         unsigned long int_fld[_NUM_FLD];
1098         char *str_fld[_NUM_FLD];
1099         int i;
1100         unsigned size;
1101
1102         nb_lcore_params = 0;
1103
1104         while ((p = strchr(p0,'(')) != NULL) {
1105                 ++p;
1106                 if((p0 = strchr(p,')')) == NULL)
1107                         return -1;
1108
1109                 size = p0 - p;
1110                 if(size >= sizeof(s))
1111                         return -1;
1112
1113                 snprintf(s, sizeof(s), "%.*s", size, p);
1114                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1115                                                                 _NUM_FLD)
1116                         return -1;
1117                 for (i = 0; i < _NUM_FLD; i++){
1118                         errno = 0;
1119                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1120                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1121                                                                         255)
1122                                 return -1;
1123                 }
1124                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1125                         printf("exceeded max number of lcore params: %hu\n",
1126                                 nb_lcore_params);
1127                         return -1;
1128                 }
1129                 lcore_params_array[nb_lcore_params].port_id =
1130                                 (uint8_t)int_fld[FLD_PORT];
1131                 lcore_params_array[nb_lcore_params].queue_id =
1132                                 (uint8_t)int_fld[FLD_QUEUE];
1133                 lcore_params_array[nb_lcore_params].lcore_id =
1134                                 (uint8_t)int_fld[FLD_LCORE];
1135                 ++nb_lcore_params;
1136         }
1137         lcore_params = lcore_params_array;
1138
1139         return 0;
1140 }
1141
1142 /* Parse the argument given in the command line of the application */
1143 static int
1144 parse_args(int argc, char **argv)
1145 {
1146         int opt, ret;
1147         char **argvopt;
1148         int option_index;
1149         char *prgname = argv[0];
1150         static struct option lgopts[] = {
1151                 {"config", 1, 0, 0},
1152                 {"no-numa", 0, 0, 0},
1153                 {"enable-jumbo", 0, 0, 0},
1154                 {NULL, 0, 0, 0}
1155         };
1156
1157         argvopt = argv;
1158
1159         while ((opt = getopt_long(argc, argvopt, "p:P",
1160                                 lgopts, &option_index)) != EOF) {
1161
1162                 switch (opt) {
1163                 /* portmask */
1164                 case 'p':
1165                         enabled_port_mask = parse_portmask(optarg);
1166                         if (enabled_port_mask == 0) {
1167                                 printf("invalid portmask\n");
1168                                 print_usage(prgname);
1169                                 return -1;
1170                         }
1171                         break;
1172                 case 'P':
1173                         printf("Promiscuous mode selected\n");
1174                         promiscuous_on = 1;
1175                         break;
1176
1177                 /* long options */
1178                 case 0:
1179                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1180                                 ret = parse_config(optarg);
1181                                 if (ret) {
1182                                         printf("invalid config\n");
1183                                         print_usage(prgname);
1184                                         return -1;
1185                                 }
1186                         }
1187
1188                         if (!strncmp(lgopts[option_index].name,
1189                                                 "no-numa", 7)) {
1190                                 printf("numa is disabled \n");
1191                                 numa_on = 0;
1192                         }
1193
1194                         if (!strncmp(lgopts[option_index].name,
1195                                         "enable-jumbo", 12)) {
1196                                 struct option lenopts =
1197                                         {"max-pkt-len", required_argument, \
1198                                                                         0, 0};
1199
1200                                 printf("jumbo frame is enabled \n");
1201                                 port_conf.rxmode.jumbo_frame = 1;
1202
1203                                 /**
1204                                  * if no max-pkt-len set, use the default value
1205                                  * ETHER_MAX_LEN
1206                                  */
1207                                 if (0 == getopt_long(argc, argvopt, "",
1208                                                 &lenopts, &option_index)) {
1209                                         ret = parse_max_pkt_len(optarg);
1210                                         if ((ret < 64) ||
1211                                                 (ret > MAX_JUMBO_PKT_LEN)){
1212                                                 printf("invalid packet "
1213                                                                 "length\n");
1214                                                 print_usage(prgname);
1215                                                 return -1;
1216                                         }
1217                                         port_conf.rxmode.max_rx_pkt_len = ret;
1218                                 }
1219                                 printf("set jumbo frame "
1220                                         "max packet length to %u\n",
1221                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1222                         }
1223
1224                         break;
1225
1226                 default:
1227                         print_usage(prgname);
1228                         return -1;
1229                 }
1230         }
1231
1232         if (optind >= 0)
1233                 argv[optind-1] = prgname;
1234
1235         ret = optind-1;
1236         optind = 0; /* reset getopt lib */
1237         return ret;
1238 }
1239
1240 static void
1241 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1242 {
1243         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1244                 eth_addr->addr_bytes[0],
1245                 eth_addr->addr_bytes[1],
1246                 eth_addr->addr_bytes[2],
1247                 eth_addr->addr_bytes[3],
1248                 eth_addr->addr_bytes[4],
1249                 eth_addr->addr_bytes[5]);
1250 }
1251
1252 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1253 static void
1254 setup_hash(int socketid)
1255 {
1256         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1257                 .name = NULL,
1258                 .entries = L3FWD_HASH_ENTRIES,
1259                 .bucket_entries = 4,
1260                 .key_len = sizeof(struct ipv4_5tuple),
1261                 .hash_func = DEFAULT_HASH_FUNC,
1262                 .hash_func_init_val = 0,
1263         };
1264
1265         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1266                 .name = NULL,
1267                 .entries = L3FWD_HASH_ENTRIES,
1268                 .bucket_entries = 4,
1269                 .key_len = sizeof(struct ipv6_5tuple),
1270                 .hash_func = DEFAULT_HASH_FUNC,
1271                 .hash_func_init_val = 0,
1272         };
1273
1274         unsigned i;
1275         int ret;
1276         char s[64];
1277
1278         /* create ipv4 hash */
1279         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1280         ipv4_l3fwd_hash_params.name = s;
1281         ipv4_l3fwd_hash_params.socket_id = socketid;
1282         ipv4_l3fwd_lookup_struct[socketid] =
1283                 rte_hash_create(&ipv4_l3fwd_hash_params);
1284         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1285                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1286                                 "socket %d\n", socketid);
1287
1288         /* create ipv6 hash */
1289         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1290         ipv6_l3fwd_hash_params.name = s;
1291         ipv6_l3fwd_hash_params.socket_id = socketid;
1292         ipv6_l3fwd_lookup_struct[socketid] =
1293                 rte_hash_create(&ipv6_l3fwd_hash_params);
1294         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1295                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1296                                 "socket %d\n", socketid);
1297
1298
1299         /* populate the ipv4 hash */
1300         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1301                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1302                                 (void *) &ipv4_l3fwd_route_array[i].key);
1303                 if (ret < 0) {
1304                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1305                                 "l3fwd hash on socket %d\n", i, socketid);
1306                 }
1307                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1308                 printf("Hash: Adding key\n");
1309                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1310         }
1311
1312         /* populate the ipv6 hash */
1313         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1314                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1315                                 (void *) &ipv6_l3fwd_route_array[i].key);
1316                 if (ret < 0) {
1317                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1318                                 "l3fwd hash on socket %d\n", i, socketid);
1319                 }
1320                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1321                 printf("Hash: Adding key\n");
1322                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1323         }
1324 }
1325 #endif
1326
1327 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1328 static void
1329 setup_lpm(int socketid)
1330 {
1331         unsigned i;
1332         int ret;
1333         char s[64];
1334
1335         /* create the LPM table */
1336         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1337         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1338                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1339         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1340                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1341                                 " on socket %d\n", socketid);
1342
1343         /* populate the LPM table */
1344         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1345                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1346                         ipv4_l3fwd_route_array[i].ip,
1347                         ipv4_l3fwd_route_array[i].depth,
1348                         ipv4_l3fwd_route_array[i].if_out);
1349
1350                 if (ret < 0) {
1351                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1352                                 "l3fwd LPM table on socket %d\n",
1353                                 i, socketid);
1354                 }
1355
1356                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1357                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1358                         ipv4_l3fwd_route_array[i].depth,
1359                         ipv4_l3fwd_route_array[i].if_out);
1360         }
1361 }
1362 #endif
1363
1364 static int
1365 init_mem(unsigned nb_mbuf)
1366 {
1367         struct lcore_conf *qconf;
1368         int socketid;
1369         unsigned lcore_id;
1370         char s[64];
1371
1372         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1373                 if (rte_lcore_is_enabled(lcore_id) == 0)
1374                         continue;
1375
1376                 if (numa_on)
1377                         socketid = rte_lcore_to_socket_id(lcore_id);
1378                 else
1379                         socketid = 0;
1380
1381                 if (socketid >= NB_SOCKETS) {
1382                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1383                                         "out of range %d\n", socketid,
1384                                                 lcore_id, NB_SOCKETS);
1385                 }
1386                 if (pktmbuf_pool[socketid] == NULL) {
1387                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1388                         pktmbuf_pool[socketid] =
1389                                 rte_mempool_create(s, nb_mbuf,
1390                                         MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1391                                         sizeof(struct rte_pktmbuf_pool_private),
1392                                         rte_pktmbuf_pool_init, NULL,
1393                                         rte_pktmbuf_init, NULL,
1394                                         socketid, 0);
1395                         if (pktmbuf_pool[socketid] == NULL)
1396                                 rte_exit(EXIT_FAILURE,
1397                                         "Cannot init mbuf pool on socket %d\n",
1398                                                                 socketid);
1399                         else
1400                                 printf("Allocated mbuf pool on socket %d\n",
1401                                                                 socketid);
1402
1403 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1404                         setup_lpm(socketid);
1405 #else
1406                         setup_hash(socketid);
1407 #endif
1408                 }
1409                 qconf = &lcore_conf[lcore_id];
1410                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1411 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1412                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1413 #endif
1414         }
1415         return 0;
1416 }
1417
1418 /* Check the link status of all ports in up to 9s, and print them finally */
1419 static void
1420 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1421 {
1422 #define CHECK_INTERVAL 100 /* 100ms */
1423 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1424         uint8_t portid, count, all_ports_up, print_flag = 0;
1425         struct rte_eth_link link;
1426
1427         printf("\nChecking link status");
1428         fflush(stdout);
1429         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1430                 all_ports_up = 1;
1431                 for (portid = 0; portid < port_num; portid++) {
1432                         if ((port_mask & (1 << portid)) == 0)
1433                                 continue;
1434                         memset(&link, 0, sizeof(link));
1435                         rte_eth_link_get_nowait(portid, &link);
1436                         /* print link status if flag set */
1437                         if (print_flag == 1) {
1438                                 if (link.link_status)
1439                                         printf("Port %d Link Up - speed %u "
1440                                                 "Mbps - %s\n", (uint8_t)portid,
1441                                                 (unsigned)link.link_speed,
1442                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1443                                         ("full-duplex") : ("half-duplex\n"));
1444                                 else
1445                                         printf("Port %d Link Down\n",
1446                                                 (uint8_t)portid);
1447                                 continue;
1448                         }
1449                         /* clear all_ports_up flag if any link down */
1450                         if (link.link_status == 0) {
1451                                 all_ports_up = 0;
1452                                 break;
1453                         }
1454                 }
1455                 /* after finally printing all link status, get out */
1456                 if (print_flag == 1)
1457                         break;
1458
1459                 if (all_ports_up == 0) {
1460                         printf(".");
1461                         fflush(stdout);
1462                         rte_delay_ms(CHECK_INTERVAL);
1463                 }
1464
1465                 /* set the print_flag if all ports up or timeout */
1466                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1467                         print_flag = 1;
1468                         printf("done\n");
1469                 }
1470         }
1471 }
1472
1473 int
1474 MAIN(int argc, char **argv)
1475 {
1476         struct lcore_conf *qconf;
1477         struct rte_eth_dev_info dev_info;
1478         struct rte_eth_txconf *txconf;
1479         int ret;
1480         unsigned nb_ports;
1481         uint16_t queueid;
1482         unsigned lcore_id;
1483         uint64_t hz;
1484         uint32_t n_tx_queue, nb_lcores;
1485         uint8_t portid, nb_rx_queue, queue, socketid;
1486
1487         /* catch SIGINT and restore cpufreq governor to ondemand */
1488         signal(SIGINT, signal_exit_now);
1489
1490         /* init EAL */
1491         ret = rte_eal_init(argc, argv);
1492         if (ret < 0)
1493                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1494         argc -= ret;
1495         argv += ret;
1496
1497         /* init RTE timer library to be used late */
1498         rte_timer_subsystem_init();
1499
1500         /* parse application arguments (after the EAL ones) */
1501         ret = parse_args(argc, argv);
1502         if (ret < 0)
1503                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1504
1505         if (check_lcore_params() < 0)
1506                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1507
1508         ret = init_lcore_rx_queues();
1509         if (ret < 0)
1510                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1511
1512
1513         nb_ports = rte_eth_dev_count();
1514         if (nb_ports > RTE_MAX_ETHPORTS)
1515                 nb_ports = RTE_MAX_ETHPORTS;
1516
1517         if (check_port_config(nb_ports) < 0)
1518                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1519
1520         nb_lcores = rte_lcore_count();
1521
1522         /* initialize all ports */
1523         for (portid = 0; portid < nb_ports; portid++) {
1524                 /* skip ports that are not enabled */
1525                 if ((enabled_port_mask & (1 << portid)) == 0) {
1526                         printf("\nSkipping disabled port %d\n", portid);
1527                         continue;
1528                 }
1529
1530                 /* init port */
1531                 printf("Initializing port %d ... ", portid );
1532                 fflush(stdout);
1533
1534                 nb_rx_queue = get_port_n_rx_queues(portid);
1535                 n_tx_queue = nb_lcores;
1536                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1537                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1538                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1539                         nb_rx_queue, (unsigned)n_tx_queue );
1540                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1541                                         (uint16_t)n_tx_queue, &port_conf);
1542                 if (ret < 0)
1543                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1544                                         "err=%d, port=%d\n", ret, portid);
1545
1546                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1547                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1548                 printf(", ");
1549
1550                 /* init memory */
1551                 ret = init_mem(NB_MBUF);
1552                 if (ret < 0)
1553                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1554
1555                 /* init one TX queue per couple (lcore,port) */
1556                 queueid = 0;
1557                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1558                         if (rte_lcore_is_enabled(lcore_id) == 0)
1559                                 continue;
1560
1561                         if (numa_on)
1562                                 socketid = \
1563                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1564                         else
1565                                 socketid = 0;
1566
1567                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1568                         fflush(stdout);
1569
1570                         rte_eth_dev_info_get(portid, &dev_info);
1571                         txconf = &dev_info.default_txconf;
1572                         if (port_conf.rxmode.jumbo_frame)
1573                                 txconf->txq_flags = 0;
1574                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1575                                                      socketid, txconf);
1576                         if (ret < 0)
1577                                 rte_exit(EXIT_FAILURE,
1578                                         "rte_eth_tx_queue_setup: err=%d, "
1579                                                 "port=%d\n", ret, portid);
1580
1581                         qconf = &lcore_conf[lcore_id];
1582                         qconf->tx_queue_id[portid] = queueid;
1583                         queueid++;
1584                 }
1585                 printf("\n");
1586         }
1587
1588         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1589                 if (rte_lcore_is_enabled(lcore_id) == 0)
1590                         continue;
1591
1592                 /* init power management library */
1593                 ret = rte_power_init(lcore_id);
1594                 if (ret)
1595                         rte_exit(EXIT_FAILURE, "Power management library "
1596                                 "initialization failed on core%u\n", lcore_id);
1597
1598                 /* init timer structures for each enabled lcore */
1599                 rte_timer_init(&power_timers[lcore_id]);
1600                 hz = rte_get_timer_hz();
1601                 rte_timer_reset(&power_timers[lcore_id],
1602                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1603                                                 power_timer_cb, NULL);
1604
1605                 qconf = &lcore_conf[lcore_id];
1606                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1607                 fflush(stdout);
1608                 /* init RX queues */
1609                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1610                         portid = qconf->rx_queue_list[queue].port_id;
1611                         queueid = qconf->rx_queue_list[queue].queue_id;
1612
1613                         if (numa_on)
1614                                 socketid = \
1615                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1616                         else
1617                                 socketid = 0;
1618
1619                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1620                         fflush(stdout);
1621
1622                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1623                                 socketid, NULL,
1624                                 pktmbuf_pool[socketid]);
1625                         if (ret < 0)
1626                                 rte_exit(EXIT_FAILURE,
1627                                         "rte_eth_rx_queue_setup: err=%d, "
1628                                                 "port=%d\n", ret, portid);
1629                 }
1630         }
1631
1632         printf("\n");
1633
1634         /* start ports */
1635         for (portid = 0; portid < nb_ports; portid++) {
1636                 if ((enabled_port_mask & (1 << portid)) == 0) {
1637                         continue;
1638                 }
1639                 /* Start device */
1640                 ret = rte_eth_dev_start(portid);
1641                 if (ret < 0)
1642                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1643                                                 "port=%d\n", ret, portid);
1644
1645                 /*
1646                  * If enabled, put device in promiscuous mode.
1647                  * This allows IO forwarding mode to forward packets
1648                  * to itself through 2 cross-connected  ports of the
1649                  * target machine.
1650                  */
1651                 if (promiscuous_on)
1652                         rte_eth_promiscuous_enable(portid);
1653         }
1654
1655         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1656
1657         /* launch per-lcore init on every lcore */
1658         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1659         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1660                 if (rte_eal_wait_lcore(lcore_id) < 0)
1661                         return -1;
1662         }
1663
1664         return 0;
1665 }